Objective Enteric pathogens have been implicated in the aetiology of inflammatory bowel disease (IBD), but increased rates of stool testing of patients with unclear gastrointestinal symptoms might cause detection bias. Hence, the objective of this study was to analyse incidence rates of Crohn's disease and ulcerative colitis among patients with Salmonella- or Campylobacter-positive and negative stool tests and to study the incidence of positive and negative stool tests among patients already diagnosed with IBD.
Methods The Danish population was followed for 94.3 million person-years during 1992–2008 using national registers to identify persons with positive and negative stool tests and patients with IBD. Using Poisson regression, incidence rate ratios (IRRs) for IBD after positive or negative stool tests and, conversely, IRRs for positive and negative stool tests following IBD, were calculated.
Results IRRs for IBD were significantly high in the first year after Salmonella- or Campylobacter-positive stool tests (IRRs 5.4–9.8), and they remained moderately increased 1–10 years later (IRRs 1.6–2.2), and less so >10 years later (IRRs 0.8–1.8). However, IRRs for IBD <1 year after a negative stool test were several-fold higher (IRRs 53.2–57.5), and a decreasing incidence pattern over time was parallel to that following positive test results. Among patients with IBD, IRRs for subsequent positive and—most notably—negative stool test results were also significantly high.
Conclusion Similarities in temporal risk patterns for IBD following positive or negative stool tests indicate that the increased occurrence of Salmonella- or Campylobacter-positive results around the time of first IBD hospitalisation results from detection bias.
- cohort studies
- inflammatory bowel disease
- Salmonella, crohn's disease
Statistics from Altmetric.com
Significance of this study
What is already known about this subject?
The aetiology of inflammatory bowel disease (IBD) remains uncertain.
Enteric infections with Salmonella and Campylobacter have been suggested to play an aetiological role in development of IBD.
The risk of detection bias in relation to this observation remains to be studied.
What are the new findings?
Temporal risk patterns for IBD are similar following negative and positive stool tests.
This observation indicates that increased occurrence of Salmonella and Campylobacter around the time of diagnosis results from detection bias due to increased rates of stool testing.
How might it impact on clinical practice in the foreseeable future?
Occurrence of Salmonella or Campylobacter infection in patients with unspecific gastrointestinal symptoms compatible with IBD may represent a ‘by chance’ finding and should not exclude the patients from further clinical examination if gastrointestinal symptoms persist.
The inflammatory bowel diseases (IBDs), Crohn's disease (CD) and ulcerative colitis (UC), are chronic intestinal disorders of partly unknown aetiology.1 Although genes are known to play a role in disease development, especially in CD, genetic factors cannot explain the rapid increase in incidence of both CD and UC during the last decades,2 which most probably is due to environmental factors.1
Among environmental factors of potential aetiological importance are enteric pathogens, which have been hypothesised to trigger an already altered immune response or a defect in downregulation of the immune response, hence leading to chronic inflammation.3 Based on a few population-based studies,3–5 it has been suggested that enteric infections with pathogens such as Salmonella or Campylobacter may play an aetiological role in the development of IBD. In a case–control study based on the General Practice Research Database in the UK, the risk of IBD was increased 2.4-fold (95% CI 1.7 to 3.3) following infectious gastroenteritis (median follow-up 3.5 years), with the greatest risk observed during the first year of follow-up.3 Likewise, in a recent cohort study of a subset of the Danish population comparing 13 324 patients with Salmonella or Campylobacter with 26 648 unexposed controls, the HR for developing IBD after acute gastroenteritis was 2.9 (95% CI 2.2 to 3.9) during a mean follow-up of 7.5 years and 1.9 (95% CI 1.4 to 2.6) if the first year after infection was excluded.5 However, since a certain latency period between onset of symptoms and a definite diagnosis of IBD is common2 and since stool sampling and testing are part of the clinical evaluation programme when diagnosing IBD,6 we hypothesised that the observed increase in occurrence of acute bacterial gastroenteritis prior to IBD reflects detection bias related to repeated stool testing in patients with unclear gastrointestinal symptoms rather than aetiology.
The aim of the present study was to use population-based data from Denmark for the period 1992–2008 to further examine the risk of IBD following positive stool tests for Salmonella or Campylobacter. To assist in the evaluation of the suggested detection bias, we also studied the risk of IBD in persons whose stools had been examined but were found to be negative for bacterial pathogens, and the risk of both positive and negative stool tests in patients already diagnosed with IBD.
Materials and methods
We established a national cohort of 6.9 million persons alive and residing in Denmark between 1992 and 2008, using data from the Civil Registration System, a continuously updated national demographic database which contains information about name, sex, address, marital status, place of birth and dates of emigration, immigration and death for the population in Denmark. Each citizen has a unique 10-digit personal identification number that enables secure linkage of data between registers.
Stool tests for enteric bacterial pathogens
Patients with episodes of infection with Campylobacter, Salmonella enteritidis, Salmonella typhimurium or other non-typhoid Salmonellae were identified in the National Register of Enteric Pathogens at Statens Serum Institut, an enterprise under the Danish Ministry of Interior and Health whose aim is to prevent and control infectious diseases, biological threats and congenital disorders. In this register, all patients diagnosed with an enteric bacterial infection, as determined by positive stool culturing, are recorded. However, cases are only counted once if infected with the same specific pathogen within six consecutive months. Negative stool tests were defined as negative for the standard panel of bacterial pathogens including Campylobacter, Salmonella, Shigella and Yersinia or other bacteria additionally requested by the treating physician. During the study period, stool samples were examined at the central diagnostic laboratory at Statens Serum Institut or at one of 15 regional laboratories, depending on the clinic requesting the analysis. Characteristics of individuals who were stool tested were extracted from the register as described elsewhere.7 Data for positive stool tests were available from 1 January 1992 to 31 December 2008. Data for negative stool tests were available for the period 1 January 1995 to 31 May 2004 from the Statens Serum Institut laboratory, which had a geographical uptake area declining from ∼70% to 40% of the Danish population over the study period.
Diagnoses of IBD
Patients with IBD were identified in the Danish National Hospital Register, which contains information about all non-psychiatric inpatient hospital contacts in Denmark since 1977 and all ambulatory hospital contacts since 1995.8 Specifically, we considered as patients with IBD both inpatients and patients in ambulatory settings who had a recorded diagnosis of CD (International Classification of Diseases (ICD) 8 codes 56300–56309 or ICD-10 code group K50) or UC (ICD-8 codes 56319 or 56904 or ICD-10 code group K51) in the period 1992–2008. To avoid inclusion of patients with prevalent IBD and to minimise diagnostic ambiguity we excluded individuals registered with IBD during 1977–1991 and individuals with a diagnosis of both CD and UC from all analyses.
In all statistical analyses, the date of stool testing was defined as the date of receipt of the stool sample, whereas the date of IBD diagnosis was defined as the date of first hospital contact with IBD. In analyses of risk of IBD following positive stool tests, person-years at risk were counted from the time of the first positive stool test and until the first hospital contact with CD or UC, censoring periods after a second positive stool test in cohort members (n=2688) with more than one positive test recorded. In analyses of risk of IBD following a negative stool test, risk time was calculated from the date of the first negative test. If a positive test followed a negative test, risk time for the negative test was censored on the date of the positive test, and calculation of risk time for a positive test was started. In analyses of risk of positive stool tests in patients with IBD, person-years at risk were counted from the date of first hospital contact with IBD and until the date of the first positive stool test for Salmonella or Campylobacter. Similarly, in the analysis of risk of a negative stool test in patients with IBD, person-years were counted until the date of the first negative stool test.
The entire Danish population was followed during the period 1 January 1992 to 31 December 2008 with notice of dates for occurrence of first infection with Salmonella or Campylobacter, first negative stool test or a diagnosis of IBD, and until occurrence of the outcome of interest, emigration, death or the end of the study (31 December 2008). To estimate the RR of IBD after a positive or negative stool sample compared with the background population, we divided the time since the stool sample into 1 year intervals. By assuming piece-wise constant incidence rates with discontinuity points defined by the time-dependent covariates, we estimated the incidence rate ratios (IRRs) with 95% CIs by Poisson regression. We initially carried out several of the statistical analyses in men and women separately, which revealed similar results in the two sexes. We also included terms for statistical interaction to address possible differences in the observed temporal associations between initial stool tests and subsequent risk of IBD in men and women, and these terms for interaction were consistently statistically non-significant (p>0.05 in likelihood ratio tests). Consequently, to gain statistical power and simplicity, we combined data for men and women, and all presented IRRs therefore represent findings in the entire Danish population, adjusted for sex, age (0–4, 5–9, …75–79, 80+ years), calendar year, ethnic background (Danish vs non-Danish, as described elsewhere9), county (14 regions), population density (<10, 11–25, 26–50, 51–100, 101–500, 501–1000, 1001+ persons per km2), and terms to account for any potential interaction between sex and age and between calendar year and county.
To explore the temporal association between positive and negative stool tests and subsequent risk of IBD up to 12 years after, we made a graphical presentation of the IRR of IBD following positive or negative stool tests. Specifically, we divided the observation time and counts of IBD into 1 month intervals after the stool test, and the resulting monthly IRRs for each of the IBDs were smoothed using restricted cubic splines.10
Assuming causality, we estimated the extent to which the overall incidence of CD and UC might be attributable to culture-positive Salmonella or Campylobacter infections, by means of the population attributable risk percent (PAR%) as follows:where p(infection) denotes the prevalence of Salmonella- or Campylobacter-positive stool samples in the general Danish population, and IRR is the overall IRR estimate for CD and UC, respectively, associated with such culture-positive stool samples.
Stool testing and visualising diagnostic procedures such as bowel x-ray examinations, colonoscopies or abdominal MRI scans are part of the clinical evaluation programme upon suspicion of IBD. To evaluate the suspected detection bias further, we performed a series of supplementary analyses restricted to the geographical area that included both positive and negative stool test results for the period 1 January 1995 to 31 May 2004. In one analysis, we aimed to provide a direct comparison of IBD risk among individuals with a first negative or positive stool test. To do so, we compared proportions of individuals in the stool test-negative and -positive cohorts who had a first hospital contact with IBD within 1 year after the stool test by means of the χ2 test. In another analysis, we examined if subsequent stool test activity was influenced by whether or not individuals had a first hospital contact with IBD recorded within the first year after the first stool test. To this end, we compared the average number of new stool tests performed up to 1 year after the first stool test between individuals who had a first hospital contact with IBD in the first year after the initial stool test (regardless of whether the initial stool test was positive or negative) and individuals who had no such hospital contact with IBD in the first year, respectively, assuming a negative binomial distribution of the stool tests performed. Finally, we compared the extent to which colonoscopies, colon and small bowel x-rays, and MRI scans, were performed in the first year following negative and positive stool test results. These analyses were restricted to the geographical area with available information about both negative and positive stool test results, and were further limited to the period from 1 January 2000 to 31 May 2004, for which we had information available in the Danish National Patient Register about such visualising diagnostic procedures. Specifically, we compared the proportions of initially stool test-negative and -positive individuals who had (1) a colonoscopy or colon x-ray; (2) a small bowel x-ray; or (3) an abdominal MRI scan performed within 1 year after the stool test by means of the χ2 test.
All data management and analysis were performed with SAS v9.1 (SAS Institute, Cary, North Carolina, USA).
Overall, the Danish population contributed 94.3 million person-years between 1992 and 2008. Within the observation period, 19 251 individuals (50% women) had a positive stool test for S enteritidis, 11 322 (50% women) for S typhimurium, 11 055 (51% women) for other types of non-typhoid Salmonellae and 49 420 (48% women) for Campylobacter at an overall median age of 30.5 years. During the same period, 10 069 (58% women) were recorded with a first hospital contact for CD and 24 218 (53% women) with a first contact for UC at a median age of 44 years for CD and 37 years for UC. Demographic characteristics of the study population are presented in table 1.
Risk of CD and UC following Salmonella or Campylobacter infection
The risk of CD after an enteric non-typhoid Salmonella infection was highly increased in the first year following infection (IRR 6.6, 95% CI 4.5 to 9.8) and lower, but still significantly increased, in the period 1–10 years after (IRR 1.6, 95% CI 1.2 to 2.2). After 10 years, the risk was not significantly increased (IRR 1.5, 95% CI 0.8 to 2.8) (table 2, figure 1). Likewise, a highly increased risk of CD was observed during the first year following a Campylobacter infection (IRR 5.4, 95% CI 3.8 to 7.7), less so in the following 1–10 years (IRR 1.6, 95% CI 1.2 to 2.1) and close to unity after 10 years (IRR 0.8, 95% CI 0.3 to 2.5) (table 2, figure 1).
Similarly, the risk of developing UC after an enteric non-typhoid Salmonella infection was highly increased in the first year (IRR 9.8, 95% CI 7.9 to 12.2), lower, but still significantly increased, in the following 1–10 years (IRR 2.2, 95% CI 1.9–2.6), and remained increased after 10 years (IRR 1.8, 95% CI 1.2 to 2.6) (table 2, figure 2). Following a Campylobacter infection, the risk of UC was eightfold increased during the first year (IRR 8.2, 95% CI 6.8 to 9.9), moderately increased in the period 1–10 years after (IRR 1.6, 95% CI 1.3 to 1.9) and not significantly altered after 10 years (IRR 1.5, 95% CI 0.8 to 2.6) (table 2, figure 2).
Assuming causality, we calculated the PAR% to estimate the extent to which the overall incidence of CD and UC could be explained by culture-positive Salmonella or Campylobacter infections. Specifically, we applied the overall IRR estimates for IBD risk following culture-positive stool tests for Salmonella and Campylobacter infections in table 2 and prevalence estimates for such culture-proven infections in the general population (0.69% for Salmonella and 0.77% for Campylobacter, representing percentages of the Danish population recorded with at least one positive stool test for Salmonella or Campylobacter, respectively, within the 10 year period 1999–2008), which produced estimates of the PAR% that were consistently around 1% (1.2% and 0.7% for UC and CD, respectively, following culture-positive Salmonella infection; 1.2% and 0.9% for UC and CD, respectively, following culture-positive Campylobacter infection).
Risk of CD and UC following a negative stool test
In the areas for which we had information about both positive and negative stool tests in the period 1 January 1995 to 31 May 2004, a total of 121 898 individuals had a negative stool test and 36 403 individuals had a positive stool test for Salmonella or Campylobacter. Compared with the observed risk patterns following positive stool tests for Salmonella and Campylobacter, the risks of CD and UC were even more markedly increased soon after a negative stool test (table 2). Indeed, extreme IRRs >100 were seen for both types of IBD in the first month after a negative stool test (figures 1 and 2), with overall IRRs of 53.2 for CD and 57.5 for UC in the first year following a negative stool test (table 2). During the subsequent 1–10 years, the risk of IBD was approximately fivefold higher than in the non-tested background population, and even 10 years or more after a negative stool test, risks for CD and UC remained 2.4- to 3.0-fold increased (table 2, figures 1 and 2). In a direct comparison between individuals with initially negative and positive stool tests, those with a negative test result were 5.2 times more likely (1.35%) than those with a positive test result (0.26%) to be diagnosed with IBD within 1 year after the first stool test (χ2 test, p<0.001).
Risk of positive and negative stool tests in patients with CD and UC
As seen in table 3, the risk among patients with IBD of having a positive stool test for Salmonella or Campylobacter was 10- to 15-fold increased in the first year after first hospital contact with IBD compared with the background population. This pattern of increased risk persisted throughout the observation period and remained 1.9- to 2.7-fold increased >10 years after. A similar pattern of markedly increased risk for negative stool tests was seen in the first year after first IBD hospitalisation (28- to 34-fold increased risk) followed by a lower, yet statistically significantly, 3.1- to 4.6-fold increased risk of negative stool tests >10 years after the first hospital contact for IBD (table 3).
Diagnostic activity following initial stool tests
Regardless of the outcome of the initial stool test, individuals who had a first hospital contact with IBD <1 year after the first stool test had on average 1.22 additional stool tests performed before their first IBD hospitalisation. The corresponding average number of additional stool tests performed in the first year among those who were not diagnosed with IBD in that period was only 0.66 (p<0.001). Restricting the analyses to individuals with a first negative (n=51 336) or positive (n=21 560) stool test result between 1 January 2000 and 31 May 2004, a period when supplementary information about visualising diagnostic procedures was available in the Danish National Hospital Register, revealed that within 1 year after the first stool test, individuals with a negative test result were 4.3 times more likely than those with a positive test result to undergo colonoscopy or colon x-ray (7.61% of individuals with a negative test vs 1.79% of individuals with a positive test), 4.0 times more likely to have an x-ray of the small bowel (1.96% following a negative test vs 0.49% following a positive test) and 1.9 times more likely to have an abdominal MRI scan (0.48% following a negative test vs 0.25% following a positive test) (χ2 tests, all p values<0.001).
The present population-based study covering >94 million person-years of observation of the Danish population during 1992–2008 revealed that the risk of both CD and UC is markedly increased not only following an enteric infection with Salmonella or Campylobacter but also, and even more so, following a negative stool test. The risk decreases during the first decade after the stool test and approaches unity after 10 years following positive stool tests, but remains significantly increased after 10 years following negative stool tests. Conversely, the risk of being diagnosed with Salmonella or Campylobacter and, even more so, the risk of having a negative stool test performed is markedly increased during the first year after first hospital contact with IBD and remains significantly increased >10 years thereafter. Furthermore, during the first year following a stool examination, we observed significantly higher additional stool testing activity in individuals who were than in those who were not subsequently diagnosed with IBD and significantly higher frequencies of visualising diagnostic procedures in individuals who were initially stool test negative than in those whose stool tests were positive for Salmonella or Campylobacter.
The finding of increased risks of both CD and UC following enteric infections with Salmonella or Campylobacter confirms previous findings from other population-based studies.3 5 However, our new findings of similar, and stronger, associations following negative stool tests seriously question prior interpretations of causality between enteric infections and subsequent development of IBD. Additionally, were they causally involved, the quantitative role of culture-proven Salmonella or Campylobacter infections is likely to be very limited, considering that such infections were detected in <2% of all incident IBD cases hospitalised in Denmark during 1992–2008, and accompanying estimates of the PAR% were consistently low (∼1%).
Our finding of a 28- to 34-fold increased risk of having a negative stool test performed and a 10- to 15-fold increased risk of having a positive stool test performed in the first year after first IBD hospitalisation presumably reflects close surveillance of patients rather than a greater risk of enteric infection in patients with IBD. Interestingly, the risk of UC 1–10 years after an enteric Salmonella infection was 2.2-fold increased, whereas the risk of Salmonella infection 1–10 years after a diagnosis of UC was 2.9-fold increased. If these findings mirrored causality rather than detection bias— a bias resulting from higher stool test activity in patients with preclinical or established IBD than in persons without IBD—Salmonella would play a causal role in development of IBD of the same magnitude as IBD would play a role on subsequent risk of a Salmonella infection, which is unlikely. Furthermore, our graphic illustrations of the risks of CD and UC following enteric infections showed almost identical curves for Salmonella and Campylobacter, which further argues in favour of a non-causal association. Otherwise, the alternative, rather implausible interpretation would be that enteric bacterial infections increase the risk of IBD regardless of the specific pathogen and, hence, regardless of the specific pathogen-associated immune response. The assumption of detection bias is further supported by the fact that the true incidences of Salmonella and Campylobacter infections in the population are markedly higher than what is observed based on physician-requested stool culturing,11 which implies that the likelihood of detecting infections with these pathogens will depend on the number of stool tests performed.
Other findings argue in favour of detection bias as a plausible explanation of the temporal patterns observed. In accordance with the clinical criteria for diagnosing IBD, which include the requirement for a negative stool examination,6 individuals with negative stool tests were five times more likely than those with positive stool tests to be diagnosed with IBD within the first year following the test, and patients who were diagnosed with IBD within the first year after their first stool test were more likely than those who did not develop IBD to have had additional stool tests performed. Additionally, the markedly higher level of visualising diagnostic activities such as colonoscopies, x-rays and MRI scans in stool test-negative than in stool test-positive individuals corroborates the idea that the initial peak in IBD incidence soon after a stool test, a peak that was markedly higher for individuals with negative stool tests, reflects the influence of detection bias.
The primary strength of the present study was the assessment of an unselected, national cohort of individuals with long-term follow-up and thorough registration of culture-positive enteric infections with Salmonella and Campylobacter. Likewise, due to free access to public healthcare in Denmark and systematic registration of hospital contacts, highly reliable CD and UC diagnoses were available from both inpatient and outpatient hospital settings.12 As a further strength compared with former studies of the same topic, we included information about stool samples that were negative for a standard panel of enteric pathogens from a subset covering up to 70% of the Danish population. The representativeness of findings in this subset of the population was confirmed by finding risk estimates following positive stool tests in the entire study population and in the subset that were nearly identical (data not shown). The major strength of including patients with negative stool tests in our analyses was that they served to place our findings for patients with positive stool tests in a new perspective. By finding risk associations that were consistently stronger for patients with negative stool tests than for those with positive stool tests, our study is the first to indicate that it is not the bacterial pathogens per se, but rather the testing activity as such that is associated with subsequent IBD risk. This strongly argues in favour of a non-causal association between Salmonella and Campylobacter infections and subsequent risk of IBD.
Despite its size, population-based nature, unbiased data sources and the absence of recall bias and other limitations of case–control studies, our national cohort study has limitations that should be considered. While diagnoses of CD and UC recorded in the Danish National Patient Register have been validated and found to be both accurate and almost complete,12 we cannot tell if our findings would apply to milder cases of IBD that were not severe enough to require hospital contact during the 17 years covered by our study. However, according to unpublished data from a recent IBD incidence study from Copenhagen City and County,2 only 8.6% of patients with UC and 1.0% of patients with CD had been diagnosed outside Danish hospitals, since hospital contacts include both hospitalisations and outpatient contacts in ambulatory settings. We therefore consider it unlikely that the restriction to patients in contact with Danish hospitals would in any way explain the strongly time-dependent risk patterns observed in our study. Another limitation is that our patients with IBD were identified on their date of first hospital contact, not on the date of IBD diagnosis. Consequently, an unknown proportion of patients, mainly those who had IBD symptoms for prolonged periods before their first IBD hospitalisation, were erroneously considered ‘at risk’ of developing IBD following a positive or negative stool test. Some of these stool tests might have been taken after establishment of the IBD diagnosis in one of the relatively few gastroenterology specialist settings outside the Danish hospitals. Quantitatively, however, this minor inaccuracy has hardly affected our findings. Moreover, considering that the IRRs of IBD shortly after positive and negative stool tests and the IRRs of positive and negative stool tests shortly after first IBD hospitalisation were all markedly elevated, it is unlikely that our results would have been markedly different if dates of IBD diagnosis rather than dates of first IBD hospitalisation had been available. Another cautionary note when evaluating our results is that in our analyses of the possible impact of Salmonella and Campylobacter on subsequent IBD risk we deliberately focused on patients with only one episode of these bacterial infections. Specifically, in these analyses we censored follow-up at the time of a second infection in persons with more than one infectious episode. We did so to avoid possible distorting effects of the relatively few patients (<3% of all patients with positive stool tests for Salmonella or Campylobacter) who had records of more than one infectious episode. However, a supplementary analysis in which patients with two or more infectious episodes were kept under follow-up even after their second infectious episode produced unchanged results (data not shown). Also, we did not have information on duration of symptoms prior to collection of stool samples. Such information might have helped to differentiate between tests that were taken due to genuine enteric infections and tests performed as part of diagnostic suspicion of IBD, because episodes of acute diarrhoea are likely to be shorter. Further, some negative tests may represent viral infections, which are not routinely tested for in Denmark in cases of acute gastroenteritis. Previous studies suggest that a considerable proportion of diarrhoea episodes in Denmark, notably in children and during wintertime, may be caused by norovirus or rotavirus infections.7 13 14 In the study by Garcia Rodriguez et al, patients with gastrointestinal symptoms but negative stool tests were assumed to represent cases of viral infections, potentially explaining the similar risk of IBD among stool test-positive and -negative persons as compared with a control group without gastrointestinal symptoms.3 However, in the absence of laboratory evidence to substantiate the suspicion of a viral cause for the patients' symptoms in that study as well as in the present study, we consider it more plausible that the majority of those negative tests that were performed shortly before a diagnosis of IBD represented genuinely negative stool tests examined due to as yet undiagnosed IBD-related symptoms. Lastly, due to the registry-based nature of the present study, tobacco smoking, the only well-established environmental risk factor for IBD,15 could not be controlled for.
In conclusion, the present long-term follow-up study of the entire Danish population confirms previous findings of an increased risk of IBD following enteric infections with Salmonella or Campylobacter but also reveals that the risk pattern is strikingly parallel and even more pronounced following stool examinations with a negative outcome. With due reservations, our findings strongly suggest that the risk previously ascribed to acute bacterial gastroenteritis reflects detection bias related to increased stool testing rather than causality.
Funding The study was supported by unrestricted research grants from the (Danish) Aase and Ejnar Danielsen's Foundation and the Gangsted Foundation.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.